ID3 is a successful rule induction algorithm which is used to generate decision trees. The main weakness of this algorithm is the generation of sharp decision boundaries at every node within the tree. This paper introduces a new Fuzzy Inference Algorithm (FIA), which applies the concept of fuzzy membership functions to branches within an existing decision tree (crisp tree) in order to relax the sharp decision boundaries. It is shown that this new inference algorithm will lead to a significant improvement in the classification accuracy. The algorithm allows all the information used throughout the tree to contribute towards the decision process by combining membership grades at each node using a fuzzy inference technique.
Fuzzy Inference Algorithm, Rule Induction, Decision Trees, ID3,
Membership Functions